A method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles

Abstract

The present invention discloses a method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles, and belongs to the technical field of analysis of the causes of exposed reinforcement in bridges. The method solves the problem that the conventional detection method for exposed reinforcement in bridge main beams in the prior art has low recognition efficiency and does not conduct targeted analysis on the causes of exposed reinforcement; the present invention plans the shortest flight path of the unmanned aerial vehicle based on the parameter information and operation information of the unmanned aerial vehicle; collects the image of the bridge main beam to generate an image matrix, performs image quality enhancement on the image matrix, and obtains a new image matrix; combines the disease information, constructs a data set, establishes and trains three models, fuses the loss function, and obtains the fused loss function; uses the trained basic model as the final model, and outputs the final exposed reinforcement identification result of the main beam; analyzes the causes of exposed reinforcement in the main beam through the exposed reinforcement cause index of the main beam, and obtains the main causes of exposed reinforcement in the main beam. The present invention effectively improves the recognition performance of the basic model for exposed reinforcement in the main beam, and can be applied to bridge maintenance and inspection.

Description

Technical Field

[0001] The present invention relates to a method for analyzing the causes of exposed reinforcement in a bridge main beam, and in particular to a method for analyzing the causes of exposed reinforcement in a bridge main beam based on images collected by an unmanned aerial vehicle, and belongs to the technical field of analyzing the causes of exposed reinforcement in bridges. Background Art

[0002] During the service of a bridge, the concrete may crack or peel off due to environmental factors (such as erosion and chemical corrosion), load (overload or collision), and construction quality problems (loose pouring or improper maintenance), which may eventually expose the steel bars. The exposed steel bars will directly affect the structural strength and durability of the bridge, resulting in a decrease in bearing capacity and increased safety hazards. If not handled in a timely manner, it may aggravate structural damage, increase the difficulty and cost of maintenance, and may even cause the bridge structure to become unstable or collapse. Rapid and intelligent identification of exposed steel bars in bridge main beams is crucial to maintaining bridge safety. It can help detect problems in a timely manner and save maintenance costs. At the same time, it can optimize safety monitoring and ensure the continued safe operation of the bridge.

[0003] At present, there is a lack of research on UAV inspection paths during bridge main beam inspection, which leads to increased UAV flight time and energy waste, and also causes some bridge disease areas to not be thoroughly inspected, affecting the comprehensiveness and accuracy of disease detection. It may also increase the risk of UAV operation and increase the probability of accidents. It is of great significance to conduct research on the method of determining the shortest flight path of UAVs. First, it can improve the operating efficiency of UAVs in bridge inspection, save time and energy, and reduce wear and tear on the UAV itself, extending its service life. Secondly, accurate flight path planning can ensure that the UAV can cover all key areas, improve the comprehensiveness and accuracy of disease detection, and finally, the optimized flight path can also help reduce safety risks during flight and improve overall operational safety.

[0004] On the other hand, the analysis of the causes of exposed rebar in bridge main beams plays a vital role in ensuring the safety of bridges, formulating effective repair plans, preventing future structural problems, optimizing maintenance and inspection plans, and forming long-term monitoring strategies. By accurately analyzing the causes of exposed rebar, appropriate repair materials and technologies can be selected to prevent and control potential risks, improve the overall stability and service life of the bridge, and provide key data for bridge management to ensure long-term safety and performance benefits.

[0005] In the prior art, a patent document with a publication (announcement) number of CN109087295A discloses a visual detection method for exposed rebar in bridges based on a grayscale projection method, which utilizes the grayscale projection method to effectively locate and identify exposed rebar in bridges, and eliminates complex background interference by analyzing the grayscale value distribution in the image, thereby improving the accuracy and efficiency of detection; however, the above method does not involve a rapid identification technology for exposed rebar in main beams, which may lead to the problem of exposed rebar being ignored or delayed in processing, thereby causing increased structural safety hazards, increased maintenance costs, and potential risks of bridge structure failure. At the same time, the cause analysis of exposed rebar in main beams has not yet been considered.

[0006] Therefore, a method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by drones is needed. Summary of the invention

[0007] A brief overview of the present invention is provided below in order to provide a basic understanding of certain aspects of the present invention. It should be understood that this overview is not an exhaustive overview of the present invention. It is not intended to identify key or important parts of the present invention, nor is it intended to limit the scope of the present invention. Its purpose is merely to present certain concepts in a simplified form as a prelude to a more detailed description discussed later.

[0008] In view of this, in order to solve the problem that the traditional bridge main beam exposed rebar detection method in the prior art has low recognition efficiency and does not conduct targeted analysis of the causes of exposed rebar, the present invention provides a bridge main beam exposed rebar cause analysis method based on images collected by drones.

[0009] The technical solution is as follows: A method for analyzing the causes of exposed reinforcement of bridge main beams based on images collected by drones, comprising the following steps:

[0010] S1. Plan the shortest flight path of the drone based on the drone parameter information and operation information;

[0011] S2. Collect the bridge main beam image along the shortest flight path of the UAV to generate an image matrix, and enhance the image quality to obtain a new image matrix;

[0012] S3. construct a data set based on the new image matrix combined with disease information, input the data set into the established basic model, auxiliary model and enhanced model for training, and fuse the loss functions of the three models to obtain the fused loss function;

[0013] S4. Establish an inter-layer loss function from the enhanced model to the basic model, combine the fused loss function to obtain the total loss function, train the three models, use the trained basic model as the final model, and output the final main beam exposed reinforcement recognition result;

[0014] S5. According to the final main beam exposed reinforcement identification result, the main beam exposed reinforcement cause index is used to analyze the main beam exposed reinforcement cause and obtain the main cause of the main beam exposed reinforcement.

[0015] Furthermore, the S1 specifically includes the following steps:

[0016] S11. Determine the take-off and landing points of the drone and select the observation position of the operator;

[0017] S12. Determine the flight altitude AVh according to the focal length and sensor size of the camera carried by the drone, obtain the image coverage and resolution, and set the drone flight speed to AVv;

[0018] S13. Develop the shortest flight path based on the take-off and landing points of the drone, and determine the coordinate information of the bridge main beam and the location information of the drone shooting points;

[0019] In S13, the take-off point coordinates are AVo (AVoX, AVoY, AVoZ). On this basis, all drone shooting points are determined. The number of any drone shooting point is p, and its corresponding drone shooting point coordinates are AVp (AVpX, AVpY, AVpZ). The number set of drone shooting points is expressed as {1, 2, 3, ..., XN}, and the corresponding drone shooting point coordinate set is expressed as {AV1 (AV1X, AV1Y, AV1Z), AV2 (AV2X, AV2Y, AV2Z), ..., AVXN (AVXNX, AVXNY, AVXNZ)}, where XN is the number of drone shooting points. The adjacent distance between any drone shooting point AVp and its adjacent drone shooting point AV (p+1) is calculated. ;

[0020] Taking the take-off point AVo as the starting point, according to the distance between any drone shooting point and the take-off point , calculate the distance between all drone shooting points and the take-off point, select the three drone shooting points corresponding to the minimum distance, and randomly select them to obtain the first drone shooting point that the drone arrives at after departing from the take-off point;

[0021] After determining the first shooting point that the drone arrives at, calculate the distance between all shooting points and the first shooting point that is arrived at, screen out the three drone shooting points corresponding to the smallest distance, and randomly select them to obtain the second shooting point that the drone arrives at after departing from the first shooting point it arrives at. Repeat the above steps until all the shooting points that the drone arrives at are obtained. According to the different random selections each time, calculate the total distance TDis1 of the drone's flight under the first flight mode. Repeat the above steps, the number of repetitions is CN, and integrate to obtain the total distance set of the drone's flight under different flight modes {TDis1, TDis2, TDis3, TDis4, …, TDisCN}, arrange them in order from small to large, and select the total distances and corresponding flight routes ranked in the top TN, TN is the preset value, count the intersection information of TN flight routes, and select the flight route with the largest number of intersections to obtain the shortest flight path of the drone.

[0022] Furthermore, the step S2 specifically includes the following steps:

[0023] S21. Using the high-definition camera on the drone, collect the bridge girder image ph along the shortest flight path of the drone, and the corresponding image matrix is ​​PH;

[0024] S22. Construct a 5th-order denoising kernel function N, normalize all elements in the kernel function, and obtain the normalized denoising kernel function ;

[0025] S23. According to the normalized denoising kernel function , perform denoising kernel function processing on the image to obtain a new image matrix P to achieve image quality enhancement;

[0026] In S22, the fifth-order denoising kernel function N is expressed as:

[0027] ;

[0028] ;

[0029] Where x=1, 2, 3, 4 or 5, x is the row number of the matrix, and y=1, 2, 3, 4 or 5, y is the column number of the matrix. is the standard deviation of all elements in the kernel function, is the base of natural logarithms;

[0030] Normalized denoising kernel function It is expressed as:

[0031] ;

[0032] In S23, the new image matrix P is expressed as:

[0033] .

[0034] Furthermore, the step S3 specifically includes the following steps:

[0035] S31. Based on the new image matrix, all images containing exposed reinforcement, cracks, rust and spalling of the main beam are selected as input images;

[0036] S32. Through manual annotation, the exposed reinforcement area of ​​the main beam and other three diseases are selected in the input image, a data set is constructed, and the data set is divided into a training set and a test set in a ratio of 8:2;

[0037] S33. According to the data set, a basic model, an auxiliary model and an enhanced model are constructed, and the loss functions of the four models are calculated and integrated to obtain a fused loss function;

[0038] In S32, the exposed reinforcement area of ​​the main beam is marked as Reinforcement, and the three diseases, namely cracks, corrosion and spalling, are marked as other1, other2 and other3 respectively;

[0039] In S33, a network structure of a basic model, i.e., a deep learning model A, is constructed, wherein the deep learning model A consists of 1 input layer, 5 convolutional layers, 5 activation layers, 4 pooling layers, 1 global average pooling layer and 1 output layer;

[0040] Input the data set into the deep learning model A, calculate the weights and biases of the neural network units of the deep learning model A through the error back propagation algorithm, complete the training of the deep learning model A, and obtain the trained deep learning model A and its loss function ;

[0041] The output value of the output unit node corresponding to the deep learning model A is , respectively, are the output values ​​of the first output unit node , the output value of the second output unit node , the output value of the third output unit node and the output value of the fourth output unit node , according to the exponential function First, the output values ​​of the four output unit nodes corresponding to the deep learning model A are initially converted into the probability values ​​of the first preliminary output value and the first preliminary output value. , and then further optimize and transform it to obtain the first optimized output value and the enhanced probability value of the first optimized output value ;

[0042] The enhanced probability value using the first optimized output value And the data set, combined with the error chain derivative rule, continue to train the deep learning model A. After the training is completed, the output value of the first disease is obtained The actual disease annotations corresponding to the nodes in the output layer of the deep learning model A , the output value of the first disease They are the exposed reinforcement output values ​​of the first main beam respectively , the output value of the first crack , the first rust output value and the first peeling output value ;

[0043] Construct the first auxiliary model, namely deep learning model B, which consists of 1 input layer, 10 convolutional layers, 10 activation layers, 5 pooling layers, 1 global average pooling layer and 1 output layer;

[0044] Input the data set into the deep learning model B, calculate the weights and biases of the neural network units of the deep learning model B through the error back propagation algorithm, complete the training of the deep learning model B, and obtain the trained deep learning model B and its loss function ;

[0045] The output value of the output unit node corresponding to the deep learning model B is , respectively, are the output values ​​of the fifth output unit node , the output value of the sixth output unit node , the output value of the seventh output unit node and the output value of the eighth output unit node , according to the exponential function First, the output values ​​of the four output unit nodes corresponding to the deep learning model B are initially converted into the second preliminary output value and the probability value of the second preliminary output value , and then further optimize and transform it to obtain the second optimized output value and the enhanced probability value of the second optimized output value ;

[0046] The enhanced probability value using the second optimized output value And the data set, combined with the error chain derivative rule, continue to train the deep learning model B. After the training is completed, the output value of the second disease is obtained The actual disease annotations corresponding to the nodes in the output layer of the deep learning model B , the output value of the second disease They are the exposed reinforcement output values ​​of the second main beam respectively , the output value of the second crack , Second rust output value and the second peeling output value ;

[0047] Construct the second auxiliary model, namely the deep learning model C, which consists of 1 input layer, 15 convolutional layers, 15 activation layers, 5 pooling layers, 1 global average pooling layer and 1 output layer;

[0048] Input the data set into the deep learning model C, calculate the weights and biases of the neural network units of the deep learning model C through the error back propagation algorithm, complete the training of the deep learning model C, and obtain the trained deep learning model C and its loss function ;

[0049] The output value of the output unit node corresponding to the deep learning model C is , respectively, are the output values ​​of the ninth output unit node , the output value of the tenth output unit node , the output value of the eleventh output unit node and the output value of the twelfth output unit node , according to the exponential function First, the output values ​​of the four output unit nodes corresponding to the deep learning model C are initially converted into the third preliminary output value and the probability value of the third preliminary output value , and then further optimize and transform it to obtain the third optimized output value and the enhanced probability value of the third optimized output value ;

[0050] The enhanced probability value using the third optimized output value And the data set, combined with the error chain derivative rule, continue to train the deep learning model C. After the training is completed, the output value of the third disease is obtained The actual disease annotations corresponding to the nodes in the output layer of the deep learning model B , the third disease output value They are the exposed reinforcement output values ​​of the third main beam respectively , the output value of the third crack , the third rust output value and the third peeling output value ;

[0051] Construct a reinforcement model, namely a deep learning model D, which consists of 1 input layer, 19 convolutional layers, 19 activation layers, 5 pooling layers, 1 global average pooling layer and 1 output layer;

[0052] Input the data set into the deep learning model D, calculate the weights and biases of the neural network units of the deep learning model D through the error back propagation algorithm, complete the training of the deep learning model D, and obtain the trained deep learning model D and its loss function ;

[0053] The output value of the output unit node corresponding to the deep learning model D is , respectively, are the output values ​​of the thirteenth output unit node , the output value of the fourteenth output unit node , the output value of the fifteenth output unit node and the output value of the sixteenth output unit node , according to the exponential function First, the output values ​​of the four output unit nodes corresponding to the deep learning model D are initially converted into the fourth preliminary output value and the probability value of the fourth preliminary output value , and then further optimize and transform it to obtain the fourth optimized output value and the enhanced probability value of the fourth optimized output value ;

[0054] The enhanced probability value using the fourth optimized output value And the data set, combined with the error chain derivative rule, continue to train the deep learning model D. After the training is completed, the output value of the fourth disease is obtained The actual disease annotations corresponding to the nodes in the output layer of the deep learning model D , the output value of the fourth disease They are the exposed reinforcement output values ​​of the fourth main beam respectively , the output value of the fourth crack , the fourth rust output value and the fourth peeling output value ;

[0055] According to the second disease output value is and the enhanced probability value of the first optimized output value , construct the loss function between deep learning model B and deep learning model A , get the corresponding weight coefficient;

[0056] Probability value based on the first preliminary output value The actual disease annotations corresponding to the nodes in the output layer of the deep learning model A , construct the probability value considering the first preliminary output value The loss function of the deep learning model A is , get the corresponding weight coefficient;

[0057] Based on the loss function between deep learning model B and deep learning model A and the probability value considering the first preliminary output value The loss function of the deep learning model A is , establish the relationship loss function between the output layer of deep learning model B and deep learning model A ;

[0058] According to the above operations, the relationship loss function between the output layers of deep learning model C and deep learning model B is established. , the relationship loss function between the output layer of deep learning model C and deep learning model A , the relationship loss function between the output layer of deep learning model D and deep learning model A The relationship loss function between the output layers of deep learning model D and deep learning model C .

[0059] Furthermore, the step S4 specifically includes the following steps:

[0060] S41. Establish an inter-layer loss function from the enhanced model to the basic model, and combine the fused loss function to obtain the total loss function;

[0061] S42. Based on the chain derivation rule, the three models are trained by error back propagation to obtain the trained deep learning model A, i.e., the optimal model;

[0062] In the S41, the selected layers of the deep learning model A are connected to the selected layers of the deep learning model D, the selected layers in the deep learning model A include the second layer: convolution layer 1, the third layer: activation layer 1, the fourth layer: pooling layer 1 and the seventeenth layer: output layer, and the selected layers of the deep learning model D include the fourth layer: convolution layer 2, the sixth layer: pooling layer 1, the tenth layer: activation layer 4 and the forty-sixth layer: output layer;

[0063] The selected layers of the deep learning model A are , the corresponding size is ,in, =2, 3, 4 and 17, is the selected number of layers of the deep learning model A, is the selected layer width of the deep learning model A, is the selected layer length of the deep learning model A, is the number of channels of the selected layer of deep learning model A, and the selected layer of deep learning model D is , the corresponding size is ,in, =4, 6, 10 and 46, is the selected number of layers of the deep learning model D, is the selected layer width of the deep learning model D, is the selected layer length of the deep learning model D, is the number of selected layer channels of the deep learning model D;

[0064] when When , a convolution operation is added to the selected layer output of the deep learning model A, with a convolution kernel size of 1×1 and a number of convolution kernels. , get the selected layer output of deep learning model A , and the corresponding output value of each channel is , =1, 2, 3, ..., , Output channel number for the selected layer of deep learning model A and the selected layer of deep learning model D The corresponding output values ​​of each channel are , =1, 2, 3, ..., , Output channel number for the selected layer of deep learning model D, and establish the loss function between the selected layer of deep learning model A and the selected layer of deep learning model D ;

[0065] when When , a convolution operation is added to the selected layer output of the deep learning model D, with a convolution kernel size of 1×1 and a number of convolution kernels. , the selected layer output of the deep learning model D is , and the corresponding output value of each channel is , the selected layers of the deep learning model D The corresponding output values ​​of each channel are , establish the loss function between the selected layer of deep learning model A and the selected layer of deep learning model D ;

[0066] According to the above calculation, the loss function between the second layer of deep learning model A and the fourth layer of deep learning model D is obtained , the loss function between the third layer of deep learning model A and the sixth layer of deep learning model D And the loss function between the fourth layer of deep learning model A and the tenth layer of deep learning model D , combining the relationship loss function between the output layers of deep learning model B and deep learning model A , the relationship loss function between the output layers of deep learning model C and deep learning model B , the relationship loss function between the output layer of deep learning model C and deep learning model A , the relationship loss function between the output layer of deep learning model D and deep learning model A The relationship loss function between the output layers of deep learning model D and deep learning model C , define the total loss function .

[0067] Further, in said S5, the main beam exposed reinforcement cause index includes the thickness of the concrete cover layer, the compressive strength of the concrete, the depth of steel bar corrosion, the chloride ion content, the steel bar protection layer deviation, the crack width, the concrete stress value and the number of maintenance records;

[0068] Eight types of main beam exposed reinforcement cause index data were collected respectively, and the corresponding data vectors were constructed to obtain the data vector FG corresponding to the concrete cover thickness, the data vector KY corresponding to the concrete compressive strength, the data vector XS corresponding to the steel corrosion depth, the data vector ZL corresponding to the chloride ion content, the data vector BH corresponding to the steel cover deviation, the data vector LK corresponding to the crack width, the data vector YL corresponding to the concrete stress value and the data vector WC corresponding to the number of maintenance records. All data vectors were normalized to obtain the corresponding normalized vectors, and the covariance values ​​between the main beam exposed reinforcement cause indexes were calculated, and the covariance matrix was obtained by integration. ;

[0069] Covariance matrix It is expressed as:

[0070] ;

[0071] in, are the covariance values ​​between the thickness of the concrete cover and the thickness of the concrete cover, the compressive strength of the concrete, the depth of steel corrosion, the chloride ion content, the deviation of the steel cover, the crack width, the concrete stress value, and the number of maintenance records. They are the covariance values ​​between concrete compressive strength and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number. They are the covariance values ​​between the steel bar corrosion depth and the concrete cover thickness, concrete compressive strength, steel bar corrosion depth, chloride ion content, steel bar cover deviation, crack width, concrete stress value, and maintenance record number. are the covariance values ​​between chloride ion content and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number. They are the covariance values ​​between the steel bar cover deviation and the concrete cover thickness, concrete compressive strength, steel bar corrosion depth, chloride ion content, steel bar cover deviation, crack width, concrete stress value, and maintenance record number. They are the covariance values ​​between crack width and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number. are the covariance values ​​between concrete stress value and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number. They are the covariance values ​​between the number of maintenance records and the thickness of concrete cover, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and the number of maintenance records;

[0072] According to the covariance matrix , construct the characteristic equation ,in, are the characteristic values ​​corresponding to the eight main beam exposed reinforcement cause indicators, =1, 2, 3, …, 8, is the covariance matrix The identity matrix of the same order, represents the determinant;

[0073] Solving the characteristic equation , and obtain the characteristic values ​​corresponding to the eight main beam exposed reinforcement cause indicators - , arrange them in order from largest to smallest, and label them as - , where the eigenvalue corresponding to the first main beam exposed reinforcement cause index after sorting is The corresponding main beam exposed reinforcement cause index is the factor that has the greatest impact on the main beam exposed reinforcement;

[0074] Define the threshold YZ and calculate ,when When the calculation is stopped, the characteristic value corresponding to the main beam exposed reinforcement cause index is The corresponding main beam exposed reinforcement cause index is the main cause of the main beam exposed reinforcement, among which, It is the number of the characteristic value corresponding to the sorted main beam exposed reinforcement cause index.

[0075] Furthermore, in S33, the layers in the network structure of the deep learning model A are connected in the following order:

[0076] Input layer: image size 227×227×3;

[0077] Convolution layer 1: kernel size 3×3, number of kernels 32, Stride: 1, Padding: Same;

[0078] Activation layer 1: ReLU;

[0079] Pooling layer 1: Type: Max pooling; Pooling size: 2×2; Stride: 2;

[0080] Convolutional layer 2: convolution kernel size 3×3, number of convolution kernels 64, stride: 1, padding: same;

[0081] Activation layer 2: ReLU;

[0082] Pooling layer 2: Type: max pooling, pooling size: 2×2, stride: 2;

[0083] Convolution layer 3: convolution kernel size 3×3, number of convolution kernels 128, stride: 1, padding: same;

[0084] Activation layer 3: ReLU;

[0085] Pooling layer 3: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0086] Convolutional layer 4: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0087] Activation layer 4: ReLU;

[0088] Convolution layer 5: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0089] Activation layer 5: ReLU;

[0090] Pooling layer 4: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0091] Global average pooling layer 1;

[0092] Output layer: Type: Softmax, Number of neurons: 4.

[0093] Furthermore, in S33, the layers in the network structure of the deep learning model B are connected in the following order:

[0094] Input layer: image size 227×227×3;

[0095] Convolution layer 1: kernel size 3×3, number of kernels 32, Stride: 1, Padding: Same;

[0096] Activation layer 1: ReLU;

[0097] Convolutional layer 2: convolution kernel size 3×3, number of convolution kernels 64, stride: 1, padding: same;

[0098] Activation layer 2: ReLU;

[0099] Pooling layer 1: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0100] Convolution layer 3: convolution kernel size 3×3, number of convolution kernels 64, stride: 1, padding: same;

[0101] Activation layer 3: ReLU;

[0102] Convolutional layer 4: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same;

[0103] Activation layer 4: ReLU;

[0104] Pooling layer 2: Type: max pooling, pooling size: 2×2, stride: 2;

[0105] Convolution layer 5: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same;

[0106] Activation layer 5: ReLU;

[0107] Convolution layer 6: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same;

[0108] Activation layer 6: ReLU;

[0109] Pooling layer 3: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0110] Convolution layer 7: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0111] Activation layer 7: ReLU;

[0112] Convolution layer 8: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0113] Activation layer 8: ReLU;

[0114] Convolution layer 9: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0115] Activation layer 9: ReLU;

[0116] Pooling layer 4: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0117] Convolution layer 10: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0118] Activation layer 10: ReLU;

[0119] Pooling layer 5: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0120] Global average pooling layer 1;

[0121] Output layer: Type: Softmax, Number of neurons: 4.

[0122] Furthermore, in S33, the layers in the network structure of the deep learning model C are connected in the following order:

[0123] Input layer: image size 227×227×3;

[0124] Convolution layer 1: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same;

[0125] Activation layer 1: ReLU;

[0126] Convolutional layer 2: convolution kernel size 3×3, number of convolution kernels 64, stride: 1, padding: same;

[0127] Activation layer 2: ReLU;

[0128] Pooling layer 1: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0129] Convolution layer 3: convolution kernel size 3×3, number of convolution kernels 128, stride: 1, padding: same;

[0130] Activation layer 3: ReLU;

[0131] Convolutional layer 4: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same;

[0132] Activation layer 4: ReLU;

[0133] Pooling layer 2: Type: max pooling, pooling size: 2×2, stride: 2;

[0134] Convolution layer 5: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0135] Activation layer 5: ReLU;

[0136] Convolution layer 6: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0137] Activation layer 6: ReLU;

[0138] Convolution layer 7: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0139] Activation layer 7: ReLU;

[0140] Pooling layer 3: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0141] Convolutional layer 8: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0142] Activation layer 8: ReLU;

[0143] Convolution layer 9: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0144] Activation layer 9: ReLU;

[0145] Convolution layer 10: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0146] Activation layer 10: ReLU;

[0147] Pooling layer 4: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0148] Convolution layer 11: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0149] Activation layer 11: ReLU;

[0150] Convolution layer 12: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0151] Activation layer 12: ReLU;

[0152] Convolution layer 13: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0153] Activation layer 13: ReLU;

[0154] Convolution layer 14: kernel size 3×3, number of kernels 512, stride: 1, padding: same;

[0155] Activation layer 14: ReLU;

[0156] Convolution layer 15: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0157] Activation layer 15: ReLU;

[0158] Pooling layer 5: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0159] Global average pooling layer 1;

[0160] Output layer: Type: Softmax, Number of neurons: 4.

[0161] Furthermore, in S33, the layers in the network structure of the deep learning model D are connected in the following order:

[0162] Input layer: image size 227×227×3;

[0163] Convolution layer 1: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same;

[0164] Activation layer 1: ReLU;

[0165] Convolutional layer 2: convolution kernel size 3×3, number of convolution kernels 64, stride: 1, padding: same;

[0166] Activation layer 2: ReLU;

[0167] Pooling layer 1: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0168] Convolution layer 3: convolution kernel size 3×3, number of convolution kernels 128, stride: 1, padding: same;

[0169] Activation layer 3: ReLU;

[0170] Convolutional layer 4: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same;

[0171] Activation layer 4: ReLU;

[0172] Pooling layer 2: Type: max pooling, pooling size: 2×2, stride: 2;

[0173] Convolution layer 5: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0174] Activation layer 5: ReLU;

[0175] Convolution layer 6: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0176] Activation layer 6: ReLU;

[0177] Convolution layer 7: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0178] Activation layer 7: ReLU;

[0179] Convolution layer 8: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0180] Activation layer 8: ReLU;

[0181] Pooling layer 3: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0182] Convolution layer 9: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0183] Activation layer 9: ReLU;

[0184] Convolution layer 10: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0185] Activation layer 10: ReLU;

[0186] Convolution layer 11: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0187] Activation layer 11: ReLU;

[0188] Convolution layer 12: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0189] Activation layer 12: ReLU;

[0190] Convolution layer 13: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0191] Activation layer 13: ReLU;

[0192] Convolution layer 14: kernel size 3×3, number of kernels 512, stride: 1, padding: same;

[0193] Activation layer 14: ReLU;

[0194] Pooling layer 4: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0195] Convolution layer 15: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0196] Activation layer 15: ReLU;

[0197] Convolution layer 16: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0198] Activation layer 16: ReLU;

[0199] Convolution layer 17: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0200] Activation layer 17: ReLU;

[0201] Convolution layer 18: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0202] Activation layer 18: ReLU;

[0203] Convolution layer 19: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0204] Activation layer 19: ReLU;

[0205] Pooling layer 5: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0206] Global average pooling layer 1;

[0207] Output layer: Type: Softmax, Number of neurons: 4.

[0208] The beneficial effects of the present invention are as follows: when the present invention uses a drone to detect exposed rebar in a bridge main beam, designing the shortest flight path helps to improve detection efficiency, reduce the required time and battery energy consumption, thereby increasing the operating endurance of the drone, reducing operating costs, and being able to obtain integrity information of key structures in the shortest time, which is crucial to ensuring the safe operation of structures such as bridges; the present invention constructs a basic model, an auxiliary model, and a reinforcement model to realize intelligent identification of exposed rebar in bridge main beams, and integrates their inter-layer loss functions so that the basic model can learn the characteristics of the reinforcement model at a deeper level, so that the basic model can be better generalized, and the recognition performance of exposed rebar in the main beam is further improved; the present invention analyzes the causes of exposed rebar in the main beam, determines the causes of exposed rebar in the main beam, and can provide key data for bridge maintenance management to ensure safe operation of bridges. BRIEF DESCRIPTION OF THE DRAWINGS

[0209] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0210] Figure 1 The figure is a flow chart of a method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by drones. DETAILED DESCRIPTION

[0211] In order to make the technical solutions and advantages of the embodiments of the present invention more clearly understood, the exemplary embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than an exhaustive list of all the embodiments. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0212] refer to Figure 1 The present embodiment is described in detail. A method for analyzing the cause of exposed reinforcement of a bridge main beam based on images collected by a drone specifically comprises the following steps:

[0213] S1. Plan the shortest flight path of the drone based on the drone parameter information and operation information;

[0214] S2. Collect the bridge main beam image along the shortest flight path of the UAV to generate an image matrix, and enhance the image quality to obtain a new image matrix;

[0215] S3. construct a data set based on the new image matrix combined with disease information, input the data set into the established basic model, auxiliary model and enhanced model for training, and fuse the loss functions of the three models to obtain the fused loss function;

[0216] S4. Establish an inter-layer loss function from the enhanced model to the basic model, combine the fused loss function to obtain the total loss function, train the three models, use the trained basic model as the final model, and output the final main beam exposed reinforcement recognition result;

[0217] S5. According to the final main beam exposed reinforcement identification result, the main beam exposed reinforcement cause index is used to analyze the main beam exposed reinforcement cause and obtain the main cause of the main beam exposed reinforcement.

[0218] Furthermore, the S1 specifically includes the following steps:

[0219] S11. Determine the take-off and landing points of the drone and select the observation position of the operator;

[0220] Specifically, before the drone takes off, it is necessary to choose a location where the drone can take off and land safely, away from people and traffic, and avoid high-voltage wires and large metal structures that may interfere with the drone's navigation system. At the same time, choose an observation position with a good field of view so that the operator can maintain a good line of sight and facilitate operation;

[0221] S12. Determine the appropriate flight altitude AVh according to the focal length and sensor size of the camera carried by the drone, obtain sufficient image coverage and resolution, and set the drone flight speed to AVv;

[0222] S13. Develop a flight path based on the take-off and landing points of the drone. Based on the drawing parameters of the bridge, combined with the field of view and clarity of the camera carried by the drone, determine the coordinate information of the bridge main beam and the shooting point information of the drone. Ensure that the images of all point information can fully and clearly cover the bridge main beam, and design the shortest flight path of the drone.

[0223] In S13, the take-off point coordinates are AVo (AVoX, AVoY, AVoZ). On this basis, all drone shooting points are determined. The number of any drone shooting point is p, and its corresponding drone shooting point coordinates are AVp (AVpX, AVpY, AVpZ). The number set of drone shooting points is expressed as {1, 2, 3, ..., XN}, and the corresponding drone shooting point coordinate set is expressed as {AV1 (AV1X, AV1Y, AV1Z), AV2 (AV2X, AV2Y, AV2Z), ..., AVXN (AVXNX, AVXNY, AVXNZ)}, where XN is the number of drone shooting points. The adjacent distance between any drone shooting point AVp and its adjacent drone shooting point AV (p+1) is calculated. ;

[0224] Adjacent distance It is expressed as:

[0225] ;

[0226] Among them, (AVpX, AVpY, AVpZ) are the coordinates of any drone shooting point, and (AV(p+1)X, AV(p+1)Y, AV(p+1)Z) are the coordinates of the drone shooting point adjacent to any drone shooting point;

[0227] Taking the take-off point AVo as the starting point, according to the distance between any drone shooting point and the take-off point , calculate the distance between all drone shooting points and the take-off point, select the three drone shooting points corresponding to the minimum distance, and randomly select them to obtain the first drone shooting point that the drone arrives at after departing from the take-off point;

[0228] distance It is expressed as:

[0229] ;

[0230] After determining the first shooting point that the drone arrives at, calculate the distance between all shooting points and the first shooting point that is arrived at, screen out the three drone shooting points corresponding to the smallest distance, and randomly select them to obtain the second shooting point that the drone arrives at after departing from the first shooting point that it arrives at. It should be noted that the route of the drone will not return, that is, the drone will not fly to the same shooting point twice. Repeat the above steps until all the shooting points arrived at by the drone are obtained. According to the difference in each random selection, calculate the total flight distance TDis1 of the drone under the first flight mode, repeat the above steps, the number of repetitions is CN, and integrate to obtain the total flight distance set of the drone under different flight modes {TDis1, TDis2, TDis3, TDis4, …, TDisCN}, arrange them in order from small to large, and select the total distance and corresponding flight route ranked in the top TN, TN is a preset value, count the intersection information of TN flight routes, and select the flight route with the largest number of intersections to obtain the shortest flight path of the drone.

[0231] Furthermore, the step S2 specifically includes the following steps:

[0232] S21. Using the high-definition camera on the drone, collect the bridge girder image ph along the shortest flight path of the drone, and the corresponding image matrix is ​​PH;

[0233] S22. Construct a 5th-order denoising kernel function N, normalize all elements in the kernel function, and obtain the normalized denoising kernel function ;

[0234] S23. According to the normalized denoising kernel function , perform denoising kernel function processing on the image to obtain a new image matrix P to achieve image quality enhancement;

[0235] In S22, the fifth-order denoising kernel function N is expressed as:

[0236] ;

[0237] ;

[0238] Where x=1, 2, 3, 4 or 5, x is the row number of the matrix, and y=1, 2, 3, 4 or 5, y is the column number of the matrix. is the standard deviation of all elements in the kernel function, is the base of natural logarithms;

[0239] Normalized denoising kernel function It is expressed as:

[0240] ;

[0241] In S23, the new image matrix P is expressed as:

[0242] ;

[0243] Specifically, the bridge girder images collected by drones contain noise, which is mainly caused by environmental factors such as wind, vibration, and changes in lighting, as well as the flight stability of the drone itself and the imaging quality of the camera. These factors may cause image blur, improper exposure, or stray pixels, thus forming noise; noise has a significant impact on the identification of exposed reinforcement in bridge girders, reducing the clarity and contrast of the image, making it difficult to distinguish between exposed reinforcement and the surrounding environment, and increasing the risk of misidentification and missed detection. Noise may also interfere with the image analysis algorithm, resulting in increased computational complexity in the recognition process, affecting the accuracy and efficiency of detection, so the quality of the bridge girder images is enhanced.

[0244] Furthermore, the step S3 specifically includes the following steps:

[0245] S31. Based on the new image matrix, all images containing exposed reinforcement, cracks, rust and spalling of the main beam are selected as input images;

[0246] S32. Through manual annotation, the exposed reinforcement area of ​​the main beam and other three diseases are selected in the input image, a data set is constructed, and the data set is divided into a training set and a test set in a ratio of 8:2;

[0247] S33. According to the data set, a basic model, an auxiliary model and an enhanced model are constructed, and the loss functions of the three models are calculated and integrated to obtain a fused loss function;

[0248] In S32, the exposed reinforcement area of ​​the main beam is marked as Reinforcement, and the three diseases, namely cracks, corrosion and spalling, are marked as other1, other2 and other3 respectively;

[0249] In S33, a network structure of a basic model, i.e., a deep learning model A, is constructed, wherein the deep learning model A consists of 1 input layer, 5 convolutional layers, 5 activation layers, 4 pooling layers, 1 global average pooling layer and 1 output layer;

[0250] Input the data set into the deep learning model A, calculate the weights and biases of the neural network units of the deep learning model A through the error back propagation algorithm, complete the training of the deep learning model A, and obtain the trained deep learning model A and its loss function ;

[0251] Loss function of the trained deep learning model A It is expressed as:

[0252] ;

[0253] in, is the total number of data samples in the dataset of deep learning model A, For the The true labels of samples, For deep learning model A, predict The probability that a sample is of category 1, is the natural logarithm;

[0254] The output value of the output unit node corresponding to the deep learning model A is , respectively, are the output values ​​of the first output unit node , the output value of the second output unit node , the output value of the third output unit node and the output value of the fourth output unit node , according to the exponential function First, the output values ​​of the four output unit nodes corresponding to the deep learning model A are initially converted into the probability values ​​of the first preliminary output value and the first preliminary output value. , and then further optimize and transform it to obtain the first optimized output value and the enhanced probability value of the first optimized output value ;

[0255] Probability value of the first preliminary output value It is expressed as:

[0256] ;

[0257] The enhanced probability value of the first optimized output value It is expressed as:

[0258] ;

[0259] in, G is the enhancement coefficient;

[0260] The enhanced probability value using the first optimized output value And the data set, combined with the error chain derivative rule, continue to train the deep learning model A. After the training is completed, the output value of the first disease is obtained The actual disease annotations corresponding to the nodes in the output layer of the deep learning model A , the output value of the first disease They are the exposed reinforcement output values ​​of the first main beam respectively , the output value of the first crack , the first rust output value and the first peeling output value ;

[0261] Construct the first auxiliary model, namely deep learning model B, which consists of 1 input layer, 10 convolutional layers, 10 activation layers, 5 pooling layers, 1 global average pooling layer and 1 output layer;

[0262] Input the data set into the deep learning model B, calculate the weights and biases of the neural network units of the deep learning model B through the error back propagation algorithm, complete the training of the deep learning model B, and obtain the trained deep learning model B and its loss function ;

[0263] Loss function of the trained deep learning model B It is expressed as:

[0264] ;

[0265] in, is the total number of data samples in the dataset of deep learning model B, For the The true labels of samples, For deep learning model B to predict The probability that a sample is class 1;

[0266] The output value of the output unit node corresponding to the deep learning model B is , respectively, are the output values ​​of the fifth output unit node , the output value of the sixth output unit node , the output value of the seventh output unit node and the output value of the eighth output unit node , according to the exponential function First, the output values ​​of the four output unit nodes corresponding to the deep learning model B are initially converted into the second preliminary output value and the probability value of the second preliminary output value , and then further optimize and transform it to obtain the second optimized output value and the enhanced probability value of the second optimized output value ;

[0267] Probability value of the second preliminary output value It is expressed as:

[0268] ;

[0269] The enhanced probability value of the second optimized output value It is expressed as:

[0270] ;

[0271] The enhanced probability value using the second optimized output value And the data set, combined with the error chain derivative rule, continue to train the deep learning model B. After the training is completed, the output value of the second disease is obtained The actual disease annotations corresponding to the nodes in the output layer of the deep learning model B , the output value of the second disease They are the exposed reinforcement output values ​​of the second main beam respectively , the output value of the second crack , Second rust output value and the second peeling output value ;

[0272] Construct the second auxiliary model, namely the deep learning model C, which consists of 1 input layer, 15 convolutional layers, 15 activation layers, 5 pooling layers, 1 global average pooling layer and 1 output layer;

[0273] Input the data set into the deep learning model C, calculate the weights and biases of the neural network units of the deep learning model C through the error back propagation algorithm, complete the training of the deep learning model C, and obtain the trained deep learning model C and its loss function ;

[0274] Loss function of the trained deep learning model C It is expressed as:

[0275] ;

[0276] in, is the total number of data samples in the dataset of the deep learning model C, For the The true labels of samples, For the deep learning model C to predict The probability that a sample is class 1;

[0277] The output value of the output unit node corresponding to the deep learning model C is , respectively, are the output values ​​of the ninth output unit node , the output value of the tenth output unit node , the output value of the eleventh output unit node and the output value of the twelfth output unit node , according to the exponential function First, the output values ​​of the four output unit nodes corresponding to the deep learning model C are initially converted into the third preliminary output value and the probability value of the third preliminary output value , and then further optimize and transform it to obtain the third optimized output value and the enhanced probability value of the third optimized output value ;

[0278] The probability value of the third preliminary output value It is expressed as:

[0279] ;

[0280] The enhanced probability value of the third optimized output value It is expressed as:

[0281] ;

[0282] The enhanced probability value using the third optimized output value And the data set, combined with the error chain derivative rule, continue to train the deep learning model C. After the training is completed, the output value of the third disease is obtained The actual disease annotations corresponding to the nodes in the output layer of the deep learning model B , the third disease output value They are the exposed reinforcement output values ​​of the third main beam respectively , the output value of the third crack , the third rust output value and the third peeling output value ;

[0283] Construct a reinforcement model, namely a deep learning model D, which consists of 1 input layer, 19 convolutional layers, 19 activation layers, 5 pooling layers, 1 global average pooling layer and 1 output layer;

[0284] Input the data set into the deep learning model D, calculate the weights and biases of the neural network units of the deep learning model D through the error back propagation algorithm, complete the training of the deep learning model D, and obtain the trained deep learning model D and its loss function ;

[0285] Loss function of the trained deep learning model D It is expressed as:

[0286] ;

[0287] in, is the total number of data samples in the dataset of the deep learning model D, For the The true labels of samples, For the deep learning model D to predict The probability that a sample is class 1;

[0288] The output value of the output unit node corresponding to the deep learning model D is , respectively, are the output values ​​of the thirteenth output unit node , the output value of the fourteenth output unit node , the output value of the fifteenth output unit node and the output value of the sixteenth output unit node , according to the exponential function First, the output values ​​of the four output unit nodes corresponding to the deep learning model D are initially converted into the fourth preliminary output value and the probability value of the fourth preliminary output value , and then further optimize and transform it to obtain the fourth optimized output value and the enhanced probability value of the fourth optimized output value ;

[0289] Probability value of the second preliminary output value It is expressed as:

[0290] ;

[0291] The enhanced probability value of the second optimized output value It is expressed as:

[0292] ;

[0293] The enhanced probability value using the fourth optimized output value And the data set, combined with the error chain derivative rule, continue to train the deep learning model D. After the training is completed, the output value of the fourth disease is obtained The actual disease annotations corresponding to the nodes in the output layer of the deep learning model D , the output value of the fourth disease They are the exposed reinforcement output values ​​of the fourth main beam respectively , the output value of the fourth crack , the fourth rust output value and the fourth peeling output value ;

[0294] According to the second disease output value is and the enhanced probability value of the first optimized output value , construct the loss function between deep learning model B and deep learning model A , get the corresponding weight coefficient;

[0295] Loss function between deep learning model B and deep learning model A It is expressed as:

[0296] ;

[0297] Probability value based on the first preliminary output value The actual disease annotations corresponding to the nodes in the output layer of the deep learning model A , construct the probability value considering the first preliminary output value The loss function of the deep learning model A is , get the corresponding weight coefficient;

[0298] Consider the probability value of the first preliminary output value The loss function of the deep learning model A is It is expressed as:

[0299] ;

[0300] Based on the loss function between deep learning model B and deep learning model A and the probability value considering the first preliminary output value The loss function of the deep learning model A is , establish the relationship loss function between the output layer of deep learning model B and deep learning model A ;

[0301] The relationship loss function between the output layer of deep learning model B and deep learning model A It is expressed as:

[0302] ;

[0303] in, is the loss function between deep learning model B and deep learning model A The weight coefficient of is the probability value considering the first preliminary output value The loss function of the deep learning model A is The weight coefficient of

[0304] According to the above operations, the relationship loss function between the output layers of deep learning model C and deep learning model B is established. , the relationship loss function between the output layer of deep learning model C and deep learning model A , the relationship loss function between the output layer of deep learning model D and deep learning model A The relationship loss function between the output layers of deep learning model D and deep learning model C ;

[0305] Loss function between deep learning model C and deep learning model B It is expressed as:

[0306] ;

[0307] Consider the probability value of the second preliminary output value The loss function of the deep learning model B is It is expressed as:

[0308] ;

[0309] The relationship loss function between the output layers of deep learning model C and deep learning model B It is expressed as:

[0310] ;

[0311] in, is the loss function between deep learning model C and deep learning model B The weight coefficient of is the probability value considering the second preliminary output value The loss function of the deep learning model B is The weight coefficient of

[0312] Loss function between deep learning model C and deep learning model A It is expressed as:

[0313] ;

[0314] The relationship loss function between the output layer of deep learning model C and deep learning model A It is expressed as:

[0315] ;

[0316] in, is the loss function between deep learning model C and deep learning model A The weight coefficient of

[0317] Loss function between deep learning model D and deep learning model A It is expressed as:

[0318] ;

[0319] The relationship loss function between the output layer of deep learning model D and deep learning model A It is expressed as:

[0320] ;

[0321] in, is the loss function between deep learning model D and deep learning model A The weight coefficient of

[0322] Loss function between deep learning model D and deep learning model C It is expressed as:

[0323] ;

[0324] Consider the probability value of the third preliminary output value The loss function of the deep learning model C It is expressed as:

[0325] ;

[0326] The relationship loss function between the output layers of deep learning model D and deep learning model C It is expressed as:

[0327] ;

[0328] in, is the loss function between deep learning model D and deep learning model C The weight coefficient of

[0329] Specifically, when establishing the fast intelligent recognition model of exposed steel bars in main beams, namely the basic model, auxiliary model and reinforcement model, the images collected by the drone will show the exposed areas of steel bars, which contrast with other elements in the environment (such as cracks, spots or shadows). Exposed steel bars usually appear as irregular geometric shapes and are difficult to identify due to background noise, lighting conditions and changes in viewing angles. In order to accurately identify the features of the exposed steel bar area in the main beam, the model structure includes an input layer, a convolutional layer, an activation layer, a pooling layer, a global average pooling layer and an output layer. The functions of each structural layer are as follows:

[0330] Input layer: receives the main beam image data collected by the drone and provides the model with the original pixel value input; Convolution layer: is responsible for extracting features in the image. The convolution layer detects local features in the image, such as edges, textures, and shapes, by applying multiple filters. For exposed rebar recognition, the features include the straight lines, boundaries, and specific textures of the exposed rebar surface; Activation layer: there is usually an activation layer after each convolution layer to introduce nonlinearity, so that the model can capture more complex feature representations, which helps the model distinguish between exposed rebar and surrounding concrete with low contrast features; Pooling layer: reduces the amount of calculation by reducing the spatial resolution of the feature map and increases the model's sensitivity to small locations The pooling layer can help the model better extract key features from large amounts of data and reduce sensitivity to specific image noise and distortion. Global average pooling layer: In the deeper part of the network, the global average pooling layer takes the average value of each feature map as output, which helps to reduce the number of parameters of the model, reduces the risk of overfitting, and provides a form of spatial information compression, so that the model focuses more on detecting the presence of exposed rebar rather than its specific location. Output layer: Converts feature maps into final classification decisions. In the scenario of exposed rebar identification in main beams, the output layer predicts whether the image contains exposed rebar, or performs classification at the pixel level to provide the specific location of the exposed rebar.

[0331] Furthermore, the step S4 specifically includes the following steps:

[0332] S41. Establish an inter-layer loss function from the enhanced model to the basic model, and combine the fused loss function to obtain the total loss function;

[0333] S42. Based on the chain derivation rule, the three models are trained by error back propagation to obtain the trained deep learning model A, i.e., the optimal model;

[0334] In the S41, the selected layers of the deep learning model A are connected to the selected layers of the deep learning model D, the selected layers in the deep learning model A include the second layer: convolution layer 1, the third layer: activation layer 1, the fourth layer: pooling layer 1 and the seventeenth layer: output layer, and the selected layers of the deep learning model D include the fourth layer: convolution layer 2, the sixth layer: pooling layer 1, the tenth layer: activation layer 4 and the forty-sixth layer: output layer;

[0335] The selected layers of the deep learning model A are , the corresponding size is ,in, =2, 3, 4 and 17, is the selected number of layers of the deep learning model A, is the selected layer width of the deep learning model A, is the selected layer length of the deep learning model A, is the number of channels of the selected layer of deep learning model A, and the selected layer of deep learning model D is , the corresponding size is ,in, =4, 6, 10 and 46, is the selected number of layers of the deep learning model D, is the selected layer width of the deep learning model D, is the selected layer length of the deep learning model D, is the number of selected layer channels of the deep learning model D;

[0336] when When , a convolution operation is added to the selected layer output of the deep learning model A, with a convolution kernel size of 1×1 and a number of convolution kernels. , get the selected layer output of deep learning model A , and the corresponding output value of each channel is , =1, 2, 3, ..., , Output channel number for the selected layer of deep learning model A and the selected layer of deep learning model D The corresponding output values ​​of each channel are , =1, 2, 3, ..., , Output channel number for the selected layer of deep learning model D, and establish the loss function between the selected layer of deep learning model A and the selected layer of deep learning model D ;

[0337] when When , the loss function between the selected layer of deep learning model A and the selected layer of deep learning model D It is expressed as:

[0338] ;

[0339] when When , a convolution operation is added to the selected layer output of the deep learning model D, with a convolution kernel size of 1×1 and a number of convolution kernels. , the selected layer output of the deep learning model D is , and the corresponding output value of each channel is , the selected layers of the deep learning model D The corresponding output values ​​of each channel are , establish the loss function between the selected layer of deep learning model A and the selected layer of deep learning model D ;

[0340] when When , the loss function between the selected layer of deep learning model A and the selected layer of deep learning model D It is expressed as:

[0341] ;

[0342] According to the above calculation, the loss function between the second layer of deep learning model A and the fourth layer of deep learning model D is obtained , the loss function between the third layer of deep learning model A and the sixth layer of deep learning model D And the loss function between the fourth layer of deep learning model A and the tenth layer of deep learning model D , combining the relationship loss function between the output layers of deep learning model B and deep learning model A , the relationship loss function between the output layers of deep learning model C and deep learning model B , the relationship loss function between the output layer of deep learning model C and deep learning model A , the relationship loss function between the output layer of deep learning model D and deep learning model A The relationship loss function between the output layers of deep learning model D and deep learning model C , define the total loss function ;

[0343] Total loss function It is expressed as:

[0344] ;

[0345] in, is the loss function between the second layer of deep learning model A and the fourth layer of deep learning model D The corresponding weight coefficient is, is the loss function between the third layer of deep learning model A and the sixth layer of deep learning model D The corresponding weight coefficient is, is the loss function between the fourth layer of deep learning model A and the tenth layer of deep learning model D The corresponding weight coefficient;

[0346] Specifically, in order to enrich the feature representation capability of the basic model A, an inter-layer association relationship is established between the basic model (small model) and the enhanced model (large model), so that features at different levels in the enhanced model can be transferred to the basic model, further improving the performance of the basic model.

[0347] Further, in said S5, the main beam exposed reinforcement cause index includes the thickness of the concrete cover layer, the compressive strength of the concrete, the depth of steel bar corrosion, the chloride ion content, the steel bar protection layer deviation, the crack width, the concrete stress value and the number of maintenance records;

[0348] Eight types of main beam exposed reinforcement cause index data were collected respectively, and the corresponding data vectors were constructed to obtain the data vector FG corresponding to the concrete cover thickness, the data vector KY corresponding to the concrete compressive strength, the data vector XS corresponding to the steel corrosion depth, the data vector ZL corresponding to the chloride ion content, the data vector BH corresponding to the steel cover deviation, the data vector LK corresponding to the crack width, the data vector YL corresponding to the concrete stress value and the data vector WC corresponding to the number of maintenance records. All data vectors were normalized to obtain the corresponding normalized vectors, and the covariance values ​​between the main beam exposed reinforcement cause indexes were calculated, and the covariance matrix was obtained by integration. ;

[0349] Covariance matrix It is expressed as:

[0350] ;

[0351] in, are the covariance values ​​between the thickness of the concrete cover and the thickness of the concrete cover, the compressive strength of the concrete, the depth of steel corrosion, the chloride ion content, the deviation of the steel cover, the crack width, the concrete stress value, and the number of maintenance records. They are the covariance values ​​between concrete compressive strength and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number. They are the covariance values ​​between the steel bar corrosion depth and the concrete cover thickness, concrete compressive strength, steel bar corrosion depth, chloride ion content, steel bar cover deviation, crack width, concrete stress value, and maintenance record number. are the covariance values ​​between chloride ion content and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number. They are the covariance values ​​between the steel bar cover deviation and the concrete cover thickness, concrete compressive strength, steel bar corrosion depth, chloride ion content, steel bar cover deviation, crack width, concrete stress value, and maintenance record number. They are the covariance values ​​between crack width and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number. are the covariance values ​​between concrete stress value and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number. They are the covariance values ​​between the number of maintenance records and the thickness of concrete cover, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and the number of maintenance records;

[0352] According to the covariance matrix , construct the characteristic equation ,in, are the characteristic values ​​corresponding to the eight main beam exposed reinforcement cause indicators, =1, 2, 3, …, 8, is the covariance matrix The identity matrix of the same order, represents the determinant;

[0353] Solving the characteristic equation , and obtain the characteristic values ​​corresponding to the eight main beam exposed reinforcement cause indicators - , arrange them in order from largest to smallest, and label them as - , where the eigenvalue corresponding to the first main beam exposed reinforcement cause index after sorting is The corresponding main beam exposed reinforcement cause index is the factor that has the greatest impact on the main beam exposed reinforcement;

[0354] Define the threshold YZ and calculate ,when When the calculation is stopped, the characteristic value corresponding to the main beam exposed reinforcement cause index is The corresponding main beam exposed reinforcement cause index is the main cause of the main beam exposed reinforcement, among which, is the number of the eigenvalue corresponding to the sorted main beam exposed reinforcement cause index;

[0355] Specifically, the data vector FG corresponding to the thickness of the concrete cover is expressed as:

[0356] FG=[fg1 fg2 … fg9]

[0357] Among them, fg1-fg9 represent the concrete cover thickness values ​​at 9 measurement points respectively;

[0358] Perform a normalization operation on the data vector FG corresponding to the thickness of the concrete cover layer to obtain a first normalized vector FGg;

[0359] The first normalized vector FGg is expressed as:

[0360] FGg=[fgg1 fgg2 … fgg9]

[0361] Among them, fgg1-fgg9 represent the normalized values ​​of the concrete cover thickness at the 9 measurement points;

[0362] The data vector KY corresponding to the concrete compressive strength is expressed as:

[0363] KY=[ky1 ky2 … ky9]

[0364] Among them, ky1-ky9 represent the concrete compressive strength values ​​of 9 measurement points respectively;

[0365] Perform a normalization operation on the data vector KY corresponding to the concrete compressive strength to obtain a second normalized vector KYg;

[0366] The second normalized vector KYg is expressed as:

[0367] KYg=[kyg1 kyg2 … kyg9]

[0368] Among them, kyg1-kyg9 represent the normalized values ​​of concrete compressive strength at 9 measurement points;

[0369] The data vector XS corresponding to the steel bar corrosion depth is expressed as:

[0370] XS=[xs1 xs2 … xs9]

[0371] Among them, xs1-xs9 represent the steel bar corrosion depth values ​​of 9 measurement points respectively;

[0372] The data vector XS corresponding to the steel bar corrosion depth is normalized to obtain a third normalized vector XSg;

[0373] The third normalized vector XSg is expressed as:

[0374] XSg=[xsg1 xsg2 … xsg9]

[0375] Among them, xsg1-xsg9 represent the normalized values ​​of the steel bar corrosion depth at 9 measurement points;

[0376] The data vector ZL corresponding to the chloride ion content is expressed as:

[0377] ZL=[zl1 zl2 … zl9]

[0378] Among them, zl1-zl9 represent the chloride ion content values ​​of 9 measurement points respectively;

[0379] Performing a normalization operation on the data vector ZL corresponding to the chloride ion content to obtain a fourth normalized vector ZLg;

[0380] The fourth normalized vector ZLg is expressed as:

[0381] ZLg=[zlg1 zlg2 … zlg9]

[0382] Among them, zlg1-zlg9 represent the normalized values ​​of chloride ion content at 9 measurement points respectively;

[0383] The data vector BH corresponding to the steel bar cover deviation is expressed as:

[0384] BH=[bh1 bh2 … bh9]

[0385] Among them, bh1-bh9 represent the steel bar cover deviation values ​​of 9 measurement points respectively;

[0386] The data vector corresponding to the steel bar cover deviation is normalized to obtain a fifth normalized vector BHg;

[0387] The fifth normalized vector BHg is expressed as:

[0388] BHg=[bhg1 bhg2 … bhg9]

[0389] Among them, bhg1-bhg9 represent the normalized values ​​of the steel bar cover deviation values ​​at 9 measurement points respectively;

[0390] The data vector LK corresponding to the crack width is expressed as:

[0391] LK=[lk1 lk2 … lk9]

[0392] Among them, lk1-lk9 represent the crack width values ​​of 9 measurement points respectively;

[0393] Performing a normalization operation on the data vector LK corresponding to the crack width to obtain a sixth normalized vector LKg;

[0394] LKg=[lkg1 lkg2 … lkg9]

[0395] Among them, lkg1-lkg9 represent the normalized values ​​of the crack widths at the nine measurement points;

[0396] The data vector YL corresponding to the concrete stress value is expressed as:

[0397] YL=[yl1 yl2 … yl9]

[0398] Among them, yl1-yl9 represent the concrete stress values ​​of 9 measurement points respectively;

[0399] Perform a normalization operation on the data vector YL corresponding to the concrete stress value to obtain a seventh normalized vector YLg;

[0400] The seventh normalized vector YLg is expressed as:

[0401] YLg=[ylg1 ylg2 … ylg9]

[0402] Among them, ylg1-ylg9 represent the normalized values ​​of concrete stress values ​​at 9 measurement points respectively;

[0403] The data vector WC corresponding to the number of maintenance records is expressed as:

[0404] WC=[wc1 wc2 … wc9]

[0405] Among them, wc1-wc9 represent the maintenance record times of the 9 measurement points respectively;

[0406] Perform a normalization operation on the data vector WC corresponding to the number of maintenance records to obtain an eighth normalized vector WCg;

[0407] The eighth normalized vector WCg is expressed as:

[0408] WCg=[wcg1 wcg2 … wcg9]

[0409] Among them, wcg1-wcg9 represent the normalized values ​​of the maintenance record times of the nine measurement points;

[0410] The covariance value solution process takes the concrete cover thickness and concrete compressive strength as an example to calculate the covariance value between the concrete cover thickness and the concrete compressive strength. :

[0411] ;

[0412] ;

[0413] ;

[0414] in, represents the average value of the normalized value of the concrete cover thickness, It represents the average value of normalized concrete compressive strength.

[0415] Furthermore, in S33, the layers in the network structure of the deep learning model A are connected in the following order:

[0416] Input layer: image size 227×227×3;

[0417] Convolution layer 1: kernel size 3×3, number of kernels 32, Stride: 1, Padding: Same;

[0418] Activation layer 1: ReLU;

[0419] Pooling layer 1: Type: Max pooling; Pooling size: 2×2; Stride: 2;

[0420] Convolutional layer 2: convolution kernel size 3×3, number of convolution kernels 64, stride: 1, padding: same;

[0421] Activation layer 2: ReLU;

[0422] Pooling layer 2: Type: max pooling, pooling size: 2×2, stride: 2;

[0423] Convolution layer 3: convolution kernel size 3×3, number of convolution kernels 128, stride: 1, padding: same;

[0424] Activation layer 3: ReLU;

[0425] Pooling layer 3: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0426] Convolutional layer 4: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0427] Activation layer 4: ReLU;

[0428] Convolution layer 5: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0429] Activation layer 5: ReLU;

[0430] Pooling layer 4: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0431] Global average pooling layer 1;

[0432] Output layer: Type: Softmax, Number of neurons: 4.

[0433] Furthermore, in S33, the layers in the network structure of the deep learning model B are connected in the following order:

[0434] Input layer: image size 227×227×3;

[0435] Convolution layer 1: kernel size 3×3, number of kernels 32, Stride: 1, Padding: Same;

[0436] Activation layer 1: ReLU;

[0437] Convolutional layer 2: convolution kernel size 3×3, number of convolution kernels 64, stride: 1, padding: same;

[0438] Activation layer 2: ReLU;

[0439] Pooling layer 1: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0440] Convolution layer 3: convolution kernel size 3×3, number of convolution kernels 64, stride: 1, padding: same;

[0441] Activation layer 3: ReLU;

[0442] Convolutional layer 4: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same;

[0443] Activation layer 4: ReLU;

[0444] Pooling layer 2: Type: max pooling, pooling size: 2×2, stride: 2;

[0445] Convolution layer 5: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same;

[0446] Activation layer 5: ReLU;

[0447] Convolution layer 6: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same;

[0448] Activation layer 6: ReLU;

[0449] Pooling layer 3: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0450] Convolution layer 7: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0451] Activation layer 7: ReLU;

[0452] Convolution layer 8: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0453] Activation layer 8: ReLU;

[0454] Convolution layer 9: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0455] Activation layer 9: ReLU;

[0456] Pooling layer 4: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0457] Convolution layer 10: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0458] Activation layer 10: ReLU;

[0459] Pooling layer 5: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0460] Global average pooling layer 1;

[0461] Output layer: Type: Softmax, Number of neurons: 4.

[0462] Furthermore, in S33, the layers in the network structure of the deep learning model C are connected in the following order:

[0463] Input layer: image size 227×227×3;

[0464] Convolution layer 1: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same;

[0465] Activation layer 1: ReLU;

[0466] Convolutional layer 2: convolution kernel size 3×3, number of convolution kernels 64, stride: 1, padding: same;

[0467] Activation layer 2: ReLU;

[0468] Pooling layer 1: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0469] Convolution layer 3: convolution kernel size 3×3, number of convolution kernels 128, stride: 1, padding: same;

[0470] Activation layer 3: ReLU;

[0471] Convolutional layer 4: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same;

[0472] Activation layer 4: ReLU;

[0473] Pooling layer 2: Type: max pooling, pooling size: 2×2, stride: 2;

[0474] Convolution layer 5: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0475] Activation layer 5: ReLU;

[0476] Convolution layer 6: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0477] Activation layer 6: ReLU;

[0478] Convolution layer 7: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0479] Activation layer 7: ReLU;

[0480] Pooling layer 3: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0481] Convolutional layer 8: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0482] Activation layer 8: ReLU;

[0483] Convolution layer 9: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0484] Activation layer 9: ReLU;

[0485] Convolution layer 10: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0486] Activation layer 10: ReLU;

[0487] Pooling layer 4: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0488] Convolution layer 11: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0489] Activation layer 11: ReLU;

[0490] Convolution layer 12: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0491] Activation layer 12: ReLU;

[0492] Convolution layer 13: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0493] Activation layer 13: ReLU;

[0494] Convolution layer 14: kernel size 3×3, number of kernels 512, stride: 1, padding: same;

[0495] Activation layer 14: ReLU;

[0496] Convolution layer 15: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0497] Activation layer 15: ReLU;

[0498] Pooling layer 5: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0499] Global average pooling layer 1;

[0500] Output layer: Type: Softmax, Number of neurons: 4.

[0501] Furthermore, in S33, the layers in the network structure of the deep learning model D are connected in the following order:

[0502] Input layer: image size 227×227×3;

[0503] Convolution layer 1: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same;

[0504] Activation layer 1: ReLU;

[0505] Convolutional layer 2: convolution kernel size 3×3, number of convolution kernels 64, stride: 1, padding: same;

[0506] Activation layer 2: ReLU;

[0507] Pooling layer 1: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0508] Convolution layer 3: convolution kernel size 3×3, number of convolution kernels 128, stride: 1, padding: same;

[0509] Activation layer 3: ReLU;

[0510] Convolutional layer 4: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same;

[0511] Activation layer 4: ReLU;

[0512] Pooling layer 2: Type: max pooling, pooling size: 2×2, stride: 2;

[0513] Convolution layer 5: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0514] Activation layer 5: ReLU;

[0515] Convolution layer 6: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0516] Activation layer 6: ReLU;

[0517] Convolution layer 7: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0518] Activation layer 7: ReLU;

[0519] Convolution layer 8: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same;

[0520] Activation layer 8: ReLU;

[0521] Pooling layer 3: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0522] Convolution layer 9: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0523] Activation layer 9: ReLU;

[0524] Convolution layer 10: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0525] Activation layer 10: ReLU;

[0526] Convolution layer 11: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0527] Activation layer 11: ReLU;

[0528] Convolution layer 12: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0529] Activation layer 12: ReLU;

[0530] Convolution layer 13: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0531] Activation layer 13: ReLU;

[0532] Convolution layer 14: kernel size 3×3, number of kernels 512, stride: 1, padding: same;

[0533] Activation layer 14: ReLU;

[0534] Pooling layer 4: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0535] Convolution layer 15: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0536] Activation layer 15: ReLU;

[0537] Convolution layer 16: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0538] Activation layer 16: ReLU;

[0539] Convolution layer 17: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0540] Activation layer 17: ReLU;

[0541] Convolution layer 18: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0542] Activation layer 18: ReLU;

[0543] Convolution layer 19: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same;

[0544] Activation layer 19: ReLU;

[0545] Pooling layer 5: Type: Max pooling, pooling size: 2×2, Stride: 2;

[0546] Global average pooling layer 1;

[0547] Output layer: Type: Softmax, Number of neurons: 4.

[0548] Although the present invention has been described according to a limited number of embodiments, it will be apparent to those skilled in the art, with the benefit of the above description, that other embodiments may be envisioned within the scope of the invention thus described. In addition, it should be noted that the language used in this specification is selected primarily for readability and teaching purposes, rather than for explaining or defining the subject matter of the present invention. Therefore, many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the appended claims. The disclosure of the present invention is illustrative, not restrictive, with respect to the scope of the present invention, which is defined by the appended claims.

Claims

1. A method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles, characterized in that: The following steps are involved: S1. Plan the shortest flight path of the drone based on the drone parameter information and operation information; S2. Collect the bridge main beam image along the shortest flight path of the UAV to generate an image matrix, and enhance the image quality to obtain a new image matrix; S3. construct a data set based on the new image matrix combined with disease information, input the data set into the established basic model, the first auxiliary model, the second auxiliary model and the enhanced model for training, and fuse the loss functions of the four models to obtain the fused loss function; S4. Establish an inter-layer loss function from the enhanced model to the basic model, combine the fused loss function to obtain the total loss function, train the four models, use the trained basic model as the final model, and output the final main beam exposed reinforcement recognition result; S5. According to the final main beam exposed reinforcement identification result, the main beam exposed reinforcement cause index is used to analyze the main beam exposed reinforcement cause and obtain the main cause of the main beam exposed reinforcement.

2. The method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles according to claim 1 is characterized in that: The S1 specifically includes the following steps: S11. Determine the take-off and landing points of the drone and select the observation position of the operator; S12. Determine the flight altitude AVh according to the focal length and sensor size of the camera carried by the drone, obtain the image coverage and resolution, and set the drone flight speed to AVv; S13. Develop the shortest flight path based on the take-off and landing points of the drone, and determine the coordinate information of the bridge main beam and the location information of the drone shooting points; In S13, the coordinates of the take-off point are AVo (AVoX, AVoY, AVoZ). On this basis, all the drone shooting points are determined. The number of any drone shooting point is p, and its corresponding drone shooting point coordinates are AVp (AVpX, AVpY, AVpZ). The number set of the drone shooting points is expressed as {1, 2, 3, ..., XN}, and the corresponding drone shooting point coordinate set is expressed as {AV1 (AV1X, AV1Y, AV1Z), AV2 (AV2X, AV2Y, AV2Z), ..., AVXN (AVXNX, AVXNY, AVXNZ)}, where XN is the number of drone shooting points. The adjacent distance Dis between any drone shooting point AVp and its adjacent drone shooting point AV(p+1) is calculated. p-(p+1) ; Taking the take-off point AVo as the starting point, according to the distance Dis between any drone shooting point and the take-off point p-o , calculate the distance between all drone shooting points and the take-off point, select the three drone shooting points corresponding to the minimum distance, and randomly select them to obtain the first drone shooting point that the drone arrives at after departing from the take-off point; After determining the first shooting point that the drone arrives at, calculate the distance between all shooting points and the first shooting point that is arrived at, screen out the three drone shooting points corresponding to the smallest distance, and randomly select them to obtain the second shooting point that the drone arrives at after departing from the first shooting point it arrives at. Repeat the shooting point calculation steps until all the shooting points arrived at by the drone are obtained. According to the difference in each random selection, calculate the total flight distance TDis1 of the drone under the first flight mode, repeat the calculation steps of the total flight distance of the drone, the number of repetitions is CN, and integrate to obtain the total flight distance set of the drone under different flight modes {TDis1, TDis2, TDis3, TDis4, …, TDisCN}, arrange them in order from small to large, and select the total distances and corresponding flight routes ranked in the top TN, TN is the preset value, count the intersection information of TN flight routes, and select the flight route with the largest number of intersections to obtain the shortest flight path of the drone.

3. The method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles according to claim 2 is characterized in that: The S2 specifically includes the following steps: S21. Using the high-definition camera on the drone, collect the bridge girder image ph along the shortest flight path of the drone, and the corresponding image matrix is ​​PH; S22. Construct a 5th-order denoising kernel function N, and normalize all elements in the kernel function to obtain the normalized denoising kernel function N G ; S23. According to the normalized denoising kernel function N G , perform denoising kernel function processing on the image to obtain a new image matrix P to achieve image quality enhancement; In S22, the fifth-order denoising kernel function N is expressed as: N=[N xy ] Where x = 1, 2, 3, 4 or 5, x is the row number of the matrix, y = 1, 2, 3, 4 or 5, y is the column number of the matrix, σ is the standard deviation of all elements in the kernel function, and e is the base of the natural logarithm; Normalized denoising kernel function N G It is expressed as: In S23, the new image matrix P is expressed as: P=PH×N G 。 4. The method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles according to claim 3 is characterized in that: The S3 specifically includes the following steps: S31. Based on the new image matrix, all images containing exposed reinforcement, cracks, rust and spalling of the main beam are selected as input images; S32. Through manual annotation, the exposed reinforcement area of ​​the main beam and other three diseases are selected in the input image, a data set is constructed, and the data set is divided into a training set and a test set in a ratio of 8:2; S33. According to the data set, a basic model, a first auxiliary model, a second auxiliary model and an enhanced model are constructed, and the loss functions of the four models are calculated and integrated to obtain a fused loss function; In S32, the exposed reinforcement area of ​​the main beam is marked as Reinforcement, and the three diseases, namely cracks, corrosion and spalling, are marked as other1, other2 and other3 respectively; In S33, a network structure of a basic model, i.e., a deep learning model A, is constructed, wherein the deep learning model A consists of 1 input layer, 5 convolutional layers, 5 activation layers, 4 pooling layers, 1 global average pooling layer and 1 output layer; The data set is input into the deep learning model A, and the weights and biases of the neural network units of the deep learning model A are calculated through the error back propagation algorithm to complete the training of the deep learning model A and obtain the trained deep learning model A and its loss function La; The output value of the output unit node corresponding to the deep learning model A is a i , respectively, the output value a1 of the first output unit node, the output value a2 of the second output unit node, the output value a3 of the third output unit node, and the output value a4 of the fourth output unit node. According to the exponential function exp, the output values ​​of the four output unit nodes corresponding to the deep learning model A are initially converted into the first preliminary output value and the probability value f of the first preliminary output value. ai , and then further optimize and transform it to obtain the first optimized output value and the enhanced probability value g of the first optimized output value ai ; The enhanced probability value g of the first optimized output value is adopted ai And the data set, combined with the error chain derivative rule, continue to train the deep learning model A. After the training is completed, the output value j of the first disease is obtained ai The actual disease annotation h corresponding to the node of the output layer of the deep learning model A ai , the output value of the first disease j ai are the exposed reinforcement output values ​​j of the first main beam respectively a1 , the output value j of the first crack a2 , the first corrosion output value j a3 and the first peeling output value j a4 ; Construct the first auxiliary model, namely deep learning model B, which consists of 1 input layer, 10 convolutional layers, 10 activation layers, 5 pooling layers, 1 global average pooling layer and 1 output layer; The data set is input into the deep learning model B, and the weights and biases of the neural network units of the deep learning model B are calculated through the error back propagation algorithm, the training of the deep learning model B is completed, and the trained deep learning model B and its loss function Lb are obtained; The output value of the output unit node corresponding to the deep learning model B is b i , respectively, the output value b1 of the fifth output unit node, the output value b2 of the sixth output unit node, the output value b3 of the seventh output unit node, and the output value b4 of the eighth output unit node. According to the exponential function exp, the output values ​​of the four output unit nodes corresponding to the deep learning model B are initially converted into the second preliminary output value and the probability value f of the second preliminary output value. bi , and then further optimize and transform it to obtain the second optimized output value and the enhanced probability value g of the second optimized output value bi ; The enhanced probability value g of the second optimized output value is adopted bi And the data set, combined with the error chain derivative rule, continue to train the deep learning model B. After the training is completed, the output value j of the second disease is obtained bi The actual disease annotation h corresponding to the node of the output layer of the deep learning model B bi , the output value of the second disease j bi are the exposed reinforcement output values ​​j of the second main beam respectively b1 , the output value j of the second crack b2 , the second corrosion output value j b3 and the second peeling output value j b4 ; Construct the second auxiliary model, namely the deep learning model C, which consists of 1 input layer, 15 convolutional layers, 15 activation layers, 5 pooling layers, 1 global average pooling layer and 1 output layer; The data set is input into the deep learning model C, and the weights and biases of the neural network units of the deep learning model C are calculated through the error back propagation algorithm to complete the training of the deep learning model C and obtain the trained deep learning model C and its loss function Lc; The output value of the output unit node corresponding to the deep learning model C is c i , are the output value c1 of the ninth output unit node, the output value c2 of the tenth output unit node, the output value c3 of the eleventh output unit node, and the output value c4 of the twelfth output unit node, respectively. According to the exponential function exp, the output values ​​of the four output unit nodes corresponding to the deep learning model C are initially converted into the third preliminary output value and the probability value f of the third preliminary output value. ci , and then further optimize and transform it to obtain the third optimized output value and the enhanced probability value g of the third optimized output value ci ; The enhanced probability value g of the third optimized output value is adopted ci And the data set, combined with the error chain derivative rule, continue to train the deep learning model C. After the training is completed, the third disease output value j is obtained ci The actual disease annotation h corresponding to the node of the output layer of the deep learning model C ci , the third disease output value j ci are the exposed reinforcement output values ​​j of the third main beam respectively c1 , the output value j of the third crack c2 、The third rust output value j c3 and the third peeling output value j c4 ; Construct a reinforcement model, namely a deep learning model D, which consists of 1 input layer, 19 convolutional layers, 19 activation layers, 5 pooling layers, 1 global average pooling layer and 1 output layer; The data set is input into the deep learning model D, and the weights and biases of the neural network units of the deep learning model D are calculated through the error back propagation algorithm to complete the training of the deep learning model D and obtain the trained deep learning model D and its loss function Ld; The output value of the output unit node corresponding to the deep learning model D is d i , respectively, the output value d1 of the thirteenth output unit node, the output value d2 of the fourteenth output unit node, the output value d3 of the fifteenth output unit node, and the output value d4 of the sixteenth output unit node. According to the exponential function exp, the output values ​​of the four output unit nodes corresponding to the deep learning model D are initially converted into the fourth preliminary output value and the probability value f of the fourth preliminary output value. di , and then further optimize and transform it to obtain the fourth optimized output value and the enhanced probability value g of the fourth optimized output value di ; The enhanced probability value g of the fourth optimized output value is adopted di And the data set, combined with the error chain derivative rule, continue to train the deep learning model D. After the training is completed, the fourth disease output value j is obtained di The actual disease annotation h corresponding to the node of the output layer of the deep learning model D di , the fourth disease output value j di They are the exposed reinforcement output values ​​j of the fourth main beam respectively d1 , the output value j of the fourth crack d2 、The fourth rust output value j d3 and the fourth peeling output value j d4 ; According to the second disease output value is j bi and the enhanced probability value g of the first optimized output value ai , construct the loss function L between deep learning model B and deep learning model A BA , get the corresponding weight coefficient; The probability value f based on the first preliminary output value ai The actual disease annotation h corresponding to the node of the output layer of the deep learning model A ai , construct the probability value f considering the first preliminary output value ai The loss function L of the deep learning model A is AA , get the corresponding weight coefficient; Based on the loss function L between deep learning model B and deep learning model A BA and considering the probability value f of the first preliminary output value ai The loss function L of the deep learning model A is AA , establish the relationship loss function L between the output layer of deep learning model B and deep learning model A BAA ; According to the operation established by the relationship loss function, the relationship loss function L between the output layer of the deep learning model C and the deep learning model B is established. CBB , the relationship loss function L between the output layer of the deep learning model C and the deep learning model A CAA , the relationship loss function L between the output layer of the deep learning model D and the deep learning model A DAA The relationship loss function L between the output layers of the deep learning model D and the deep learning model C DCC .

5. The method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles according to claim 4 is characterized in that: The S4 specifically includes the following steps: S41. Establish an inter-layer loss function from the enhanced model to the basic model, and combine the fused loss function to obtain the total loss function; S42. Based on the chain derivation rule, the four models are trained by error back propagation to obtain the trained deep learning model A, i.e., the optimal model; In the S41, the selected layers of the deep learning model A are connected to the selected layers of the deep learning model D, the selected layers in the deep learning model A include the second layer of the deep learning model A: convolution layer 1, the third layer of the deep learning model A: activation layer 1, the fourth layer of the deep learning model A: pooling layer 1 and the seventeenth layer of the deep learning model A: output layer, and the selected layers of the deep learning model D include the fourth layer of the deep learning model D: convolution layer 2, the sixth layer of the deep learning model D: pooling layer 1, the tenth layer of the deep learning model D: activation layer 4 and the forty-sixth layer of the deep learning model D: output layer; The selected layer of the deep learning model A is Ma z , the corresponding size is na z × z ×qa z , where z = 2, 3, 4 and 17, z is the number of selected layers of the deep learning model A, na z is the selected layer width of the deep learning model A, pa z is the selected layer length of the deep learning model A, qa z is the number of channels of the selected layer of deep learning model A, and the selected layer of deep learning model D is Md m , the corresponding size is nd m ×pd m ×qd m , where m = 4, 6, 10 and 46, m is the number of selected layers of the deep learning model D, and nd m is the selected layer width of the deep learning model D, pd m is the selected layer length of the deep learning model D, qd m is the number of selected layer channels of the deep learning model D; When qa z <qd m When , a convolution operation is added to the selected layer output of the deep learning model A, with a convolution kernel size of 1×1 and a number of convolution kernels of qd m , get the selected layer output Ma of the deep learning model A zn , and the corresponding output value of each channel is ki=1,2,3,…,qa z , ki is the output channel number of the selected layer of deep learning model A, and Md is the selected layer of deep learning model D m The corresponding output values ​​of each channel are ki'=1,2,3,…,qd m , ki' is the output channel number of the selected layer of deep learning model D, and the loss function between the selected layer of deep learning model A and the selected layer of deep learning model D is established When qa z ≥qd m When , a convolution operation is added to the selected layer output of the deep learning model D, with a convolution kernel size of 1×1 and a number of convolution kernels of qa z , the selected layer output of the deep learning model D is obtained as Md mn , and the corresponding output value of each channel is The output value of each channel corresponding to the selected layer Md of the deep learning model D is Establish the loss function between the selected layers of deep learning model A and the selected layers of deep learning model D According to the calculation of the loss function between the selected layer of the established deep learning model A and the selected layer of the deep learning model D, the loss function between the second layer of the deep learning model A and the fourth layer of the deep learning model D is obtained. Loss function between the third layer of deep learning model A and the sixth layer of deep learning model D And the loss function between the fourth layer of deep learning model A and the tenth layer of deep learning model D Combine the relationship loss function L between the output layers of deep learning model B and deep learning model A BAA , the relationship loss function L between the output layers of deep learning model C and deep learning model B CBB , the relationship loss function L between the output layer of the deep learning model C and the deep learning model A CAA , the relationship loss function L between the output layer of the deep learning model D and the deep learning model A DAA The relationship loss function L between the output layers of the deep learning model D and the deep learning model C DCC , define the total loss function L total .

6. The method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles according to claim 5 is characterized in that: In S5, the main beam exposed reinforcement cause indicators include concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel protection layer deviation, crack width, concrete stress value and maintenance record number; Collect eight types of main beam exposed reinforcement cause index data respectively, construct corresponding data vectors, obtain data vector FG corresponding to concrete cover thickness, data vector KY corresponding to concrete compressive strength, data vector XS corresponding to steel bar corrosion depth, data vector ZL corresponding to chloride ion content, data vector BH corresponding to steel bar cover deviation, data vector LK corresponding to crack width, data vector YL corresponding to concrete stress value and data vector WC corresponding to maintenance record number, normalize all data vectors respectively, obtain corresponding normalized vectors, calculate covariance values ​​between main beam exposed reinforcement cause indexes, and integrate to obtain covariance matrix XFC; The covariance matrix XFC is expressed as: Among them, xfc 1-1 、xfc 1-2 、xfc 1-3 、xfc 1-4 、xfc 1-5 、xfc 1-6 、xfc 1-7 、xfc 1-8 are the covariance values ​​between the thickness of the concrete cover and the thickness of the concrete cover, the compressive strength of the concrete, the depth of steel corrosion, the chloride ion content, the deviation of the steel cover, the crack width, the concrete stress value, and the number of maintenance records, respectively. 2-1 、xfc 2-2 、xfc 2-3 、xfc 2-4 、xfc 2-5 、xfc 2-6 、xfc 2-7 、xfc 2-8 are the covariance values ​​between concrete compressive strength and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number, respectively. 3-1 、xfc 3-2 、xfc 3-3 、xfc 3-4 、xfc 3-5 、xfc 3-6 、xfc 3-7 、xfc 3-8 are the covariance values ​​between the steel bar corrosion depth and the concrete cover thickness, concrete compressive strength, steel bar corrosion depth, chloride ion content, steel bar cover deviation, crack width, concrete stress value, and maintenance record number, respectively. 4-1 、xfc 4-2 、xfc 4-3 、xfc 4-4 、xfc 4-5 、xfc 4-6 、xfc 4-7 、xfc 4-8 are the covariance values ​​between chloride ion content and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number, respectively. 5-1 、xfc 5-2 、xfc 5-3 、xfc 5-4 、xfc 5-5 、xfc 5-6 、xfc 5-7 、xfc 5-8 are the covariance values ​​between the steel bar protection layer deviation and the concrete cover thickness, concrete compressive strength, steel bar corrosion depth, chloride ion content, steel bar protection layer deviation, crack width, concrete stress value, and maintenance record number, respectively. 6-1 、xfc 6-2 、xfc 6-3 、xfc 6-4 、xfc 6-5 、xfc 6-6 、xfc 6-7 、xfc 6-8 are the covariance values ​​between crack width and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number, respectively. 7-1 、xfc 7-2 、xfc 7-3 、xfc 7-4 、xfc 7-5 、xfc 7-6 、xfc 7-7 、xfc 7-8 are the covariance values ​​between concrete stress value and concrete cover thickness, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and maintenance record number, respectively. 8-1 、xfc 8-2 、xfc 8-3 、xfc 8-4 、xfc 8-5 、xfc 8-6 、xfc 8-7 、xfc 8-8 They are the covariance values ​​between the number of maintenance records and the thickness of concrete cover, concrete compressive strength, steel corrosion depth, chloride ion content, steel cover deviation, crack width, concrete stress value, and the number of maintenance records; According to the covariance matrix XFC, construct the characteristic equation det(XFC-TZ si ·I)=0, where TZsi is the eigenvalue corresponding to the eight main beam exposed reinforcement cause indicators, si=1, 2, 3, …, 8, I is the unit matrix of the same order as the covariance matrix XFC, and det represents the determinant; Solve the characteristic equation det(XFC-TZ si I)=0, the eigenvalues ​​TZ1-TZ8 corresponding to the eight main beam exposed reinforcement cause indicators are obtained, and they are arranged in order from large to small, marked as TZN1-TZN8, among which the main beam exposed reinforcement cause indicator corresponding to the eigenvalue TZN1 corresponding to the first main beam exposed reinforcement cause indicator after sorting is the factor with the greatest impact on the main beam exposed reinforcement; Define the threshold YZ and calculate when When the calculation is stopped, the characteristic value TZN corresponding to the main beam exposed reinforcement cause index ni The corresponding main beam exposed reinforcement cause index is the main cause of the main beam exposed reinforcement, where ni is the number of the eigenvalue corresponding to the sorted main beam exposed reinforcement cause index.

7. The method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles according to claim 6 is characterized in that: In S33, the layers in the network structure of the deep learning model A are connected in the following order: Input layer of deep learning model A: image size 227×227×3; Convolutional layer 1 of deep learning model A: kernel size 3×3, number of kernels 32, Stride: 1, Padding: Same; Activation layer 1 of deep learning model A: ReLU; Pooling layer 1 of deep learning model A: Type: max pooling; Pooling size: 2×2; Stride: 2; Convolutional layer 2 of deep learning model A: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same; Activation layer 2 of deep learning model A: ReLU; Pooling layer 2 of deep learning model A: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolution layer 3 of deep learning model A: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same; Activation layer 3 of deep learning model A: ReLU; Pooling layer 3 of deep learning model A: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 4 of deep learning model A: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 4 of deep learning model A: ReLU; Convolutional layer 5 of deep learning model A: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 5 of deep learning model A: ReLU; Pooling layer 4 of deep learning model A: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Global average pooling layer 1 of deep learning model A; Output layer of deep learning model A: Type: Softmax, Number of neurons:

4.

8. The method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles according to claim 6 is characterized in that: In S33, the layers in the network structure of the deep learning model B are connected in the following order: Input layer of deep learning model B: image size 227×227×3; Convolutional layer 1 of deep learning model B: kernel size 3×3, number of kernels 32, Stride: 1, Padding: Same; Activation layer 1 of deep learning model B: ReLU; Convolutional layer 2 of deep learning model B: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same; Activation layer 2 of deep learning model B: ReLU; Pooling layer 1 of deep learning model B: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 3 of deep learning model B: kernel size 3×3, number of kernels 64, stride: 1, padding: same; Activation layer 3 of deep learning model B: ReLU; Convolutional layer 4 of deep learning model B: kernel size 3×3, number of kernels 64, stride: 1, padding: same; Activation layer 4 of deep learning model B: ReLU; Pooling layer 2 of deep learning model B: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 5 of deep learning model B: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same; Activation layer 5 of deep learning model B: ReLU; Convolutional layer 6 of deep learning model B: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same; Activation layer 6 of deep learning model B: ReLU; Pooling layer 3 of deep learning model B: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 7 of deep learning model B: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 7 of deep learning model B: ReLU; Convolutional layer 8 of deep learning model B: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 8 of deep learning model B: ReLU; Convolutional layer 9 of deep learning model B: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 9 of deep learning model B: ReLU; Pooling layer 4 of deep learning model B: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 10 of deep learning model B: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 10 of deep learning model B: ReLU; Pooling layer 5 of deep learning model B: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Global average pooling layer 1 of deep learning model B; Output layer of deep learning model B: Type: Softmax, Number of neurons:

4.

9. The method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles according to claim 6 is characterized in that: In S33, the layers in the network structure of the deep learning model C are connected in the following order: Input layer of deep learning model C: image size 227×227×3; Convolutional layer 1 of deep learning model C: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same; Activation layer 1 of deep learning model C: ReLU; Convolutional layer 2 of deep learning model C: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same; Activation layer 2 of deep learning model C: ReLU; Pooling layer 1 of deep learning model C: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolution layer 3 of deep learning model C: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same; Activation layer 3 of deep learning model C: ReLU; Convolutional layer 4 of deep learning model C: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same; Activation layer 4 of deep learning model C: ReLU; Pooling layer 2 of deep learning model C: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 5 of deep learning model C: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 5 of deep learning model C: ReLU; Convolutional layer 6 of deep learning model C: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 6 of deep learning model C: ReLU; Convolutional layer 7 of deep learning model C: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 7 of deep learning model C: ReLU; Pooling layer 3 of deep learning model C: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 8 of deep learning model C: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 8 of deep learning model C: ReLU; Convolutional layer 9 of deep learning model C: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 9 of deep learning model C: ReLU; Convolutional layer 10 of deep learning model C: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 10 of deep learning model C: ReLU; Pooling layer 4 of deep learning model C: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 11 of deep learning model C: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 11 of deep learning model C: ReLU; Convolutional layer 12 of deep learning model C: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 12 of deep learning model C: ReLU; Convolutional layer 13 of deep learning model C: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 13 of deep learning model C: ReLU; Convolutional layer 14 of deep learning model C: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 14 of deep learning model C: ReLU; Convolutional layer 15 of deep learning model C: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 15 of deep learning model C: ReLU; Pooling layer 5 of deep learning model C: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Global average pooling layer 1 of deep learning model C; Output layer of deep learning model C: Type: Softmax, Number of neurons:

4.

10. The method for analyzing the causes of exposed reinforcement in bridge main beams based on images collected by unmanned aerial vehicles according to claim 6, characterized in that: In S33, the layers in the network structure of the deep learning model D are connected in the following order: Input layer of deep learning model D: image size 227×227×3; Convolutional layer 1 of deep learning model D: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same; Activation layer 1 of deep learning model D: ReLU; Convolutional layer 2 of deep learning model D: kernel size 3×3, number of kernels 64, Stride: 1, Padding: Same; Activation layer 2 of deep learning model D: ReLU; Pooling layer 1 of deep learning model D: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolution layer 3 of deep learning model D: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same; Activation layer 3 of deep learning model D: ReLU; Convolutional layer 4 of deep learning model D: kernel size 3×3, number of kernels 128, Stride: 1, Padding: Same; Activation layer 4 of deep learning model D: ReLU; Pooling layer 2 of deep learning model D: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 5 of deep learning model D: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 5 of deep learning model D: ReLU; Convolutional layer 6 of deep learning model D: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 6 of deep learning model D: ReLU; Convolutional layer 7 of deep learning model D: kernel size 3×3, number of kernels 256, stride: 1, padding: same; Activation layer 7 of deep learning model D: ReLU; Convolutional layer 8 of deep learning model D: kernel size 3×3, number of kernels 256, Stride: 1, Padding: Same; Activation layer 8 of deep learning model D: ReLU; Pooling layer 3 of deep learning model D: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 9 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 9 of deep learning model D: ReLU; Convolutional layer 10 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 10 of deep learning model D: ReLU; Convolutional layer 11 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 11 of deep learning model D: ReLU; Convolutional layer 12 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 12 of deep learning model D: ReLU; Convolutional layer 13 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 13 of deep learning model D: ReLU; Convolutional layer 14 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 14 of deep learning model D: ReLU; Pooling layer 4 of deep learning model D: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Convolutional layer 15 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 15 of deep learning model D: ReLU; Convolutional layer 16 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 16 of deep learning model D: ReLU; Convolutional layer 17 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 17 of deep learning model D: ReLU; Convolutional layer 18 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 18 of deep learning model D: ReLU; Convolutional layer 19 of deep learning model D: kernel size 3×3, number of kernels 512, Stride: 1, Padding: Same; Activation layer 19 of deep learning model D: ReLU; Pooling layer 5 of deep learning model D: Type: Max Pooling, Pooling Size: 2×2, Stride: 2; Global average pooling layer 1 of deep learning model D; Output layer of deep learning model D: Type: Softmax, Number of neurons: 4.

Images